Where this data comes from, and why it is different tomorrow
This report is not a survey and not a sample. It reads the entire Kurdify corpus at the moment you load the page: {b} businesses and {r} reviews across {c} cities.
01
Listings, collected then corrected
Every business starts as a public record: a name, a category, a coordinate, a phone number. Owners then claim their listing and correct it themselves, which is the step that turns a scraped record into something accountable. A claimed listing is maintained by the person who answers the phone.
02
Reviews, written by customers
Ratings and written reviews come from the people who walked in. They are opinions, not audited facts, and this report treats them that way: it counts what people say and how often, never whether they were right.
03
It grows every day
New businesses register, new reviews arrive, owners fix details that were wrong. The counters on this page are read live rather than typed in, so every chart here is redrawn against a corpus that is larger than it was yesterday. Two people opening this page a month apart are not reading the same report.
Chapter one
Where the country actually trades
Every business on Kurdify carries a coordinate. Plotted together, the commercial geography of the region is far more concentrated than a map of the country suggests.
For example
.
Fifteen thousand businesses
Each point is one business. The shape you are looking at is not a border, it is commerce: the road network and the settled valleys of Kurdistan drawn entirely by where people opened a shop.
Erbil
The largest single concentration, and the most reviewed by a wide margin. A business here collects several times the public attention of one in a smaller town.
Sulaymaniyah
The second centre, with its own dense commercial core and a rating profile very close to Erbil's.
Duhok
The northern centre. Its customers are measurably the toughest graders in the country, which makes it the most informative city in the dataset.
The concentration
Brightness now shows density. Three cities hold roughly two thirds of every business and 85% of the reviews that can be tied to a listing. This is not one national market, it is three city markets and a long tail of towns.
Finding
Coverage and product strategy cannot be the same everywhere. The three big cities have enough reviews per business to rank and filter meaningfully. In the smaller towns the scarce resource is not ranking, it is any information at all.
Chapter two
What customers actually judge you on
Reading the text of every review that contains words, and tagging each against the themes people raise, produces a ranking that most business owners guess wrong.
For example
.
Themes by how often they appear, and what they cost
Bars show how often a theme is mentioned. The dot shows the average rating of reviews that mention it, against the corpus average.
Share of reviews mentioningRates above averageRates below average
Everything people mention
Fifteen themes, ranked by how often reviewers raise them. The dot on the right shows what each theme does to a rating.
What they talk about
Service and food dominate the conversation. But look at their dots: both sit almost exactly on the average line. They fill reviews without deciding them.
What actually moves the score
Price is mentioned far less than service, yet it pulls the furthest below the average of any high-volume theme. This single bar is the most commercially important line on the chart.
What lifts a rating
Atmosphere, local pride and hospitality all sit above the line. They are mentioned rarely, but when they are, the review is warm.
How Kurdistan rates
Distribution
Every rated review, by score. The market is strongly top-heavy.
Unhappy customers write more
Length
Average characters written, by score. Praise is a gesture; criticism is an argument.
Service and food are mentioned most, but both rate close to the overall average. They are what people talk about, not what decides the score. Price is the theme that moves the needle: it carries the lowest average of any high-volume theme in the whole corpus.
The practical reading is that people mention service when they are pleased and mention price when they are not. A customer who brings up cost in a review is, on average, already unhappy about something else.
Finding
Price is the emotional centre of this market. Not cuisine, not decor, not location. A business that communicates its prices clearly before the customer commits removes the single largest source of disappointment in the dataset.
Chapter three
What turns a bad day into a one-star review
Frequency is one thing, severity is another. Some complaints are grumbles. Others reliably end in the lowest possible score.
For example
.
One star against five, by theme
For reviews mentioning each theme, the share landing on one star and the share landing on five. Sorted by the ratio, so the themes at the top are the ones most likely to be punished publicly.
One starFive stars
The angriest themes in the corpus are not about quality at all. They are about a promise that failed. Wi-Fi that does not work, an order that arrives late, a price that turns out not to be the price. These are binary: they either hold or the review is furious.
At the other end, atmosphere and hospitality almost never produce a one-star review. A customer who disliked the decor writes three stars. A customer who felt cheated writes one.
Finding
Reliability beats excellence in this market. Being merely expensive is survivable. Being seen as dishonest is not: language about trust and cheating appears in a small share of reviews, but those reviews are drastically more likely to be one star than any other kind.
Interlude
The mind that judges
Step away from this market for a moment. Every rating anywhere in the world is produced by the same instrument: a human memory, working from a fraction of what happened, under conditions it did not choose. That instrument has known and measurable habits. None of them are flaws exactly. They are what judgement has to look like when the alternative is remembering everything, which no mind can afford.
The peak and the end
You do not remember an experience. You remember two moments of it.
Asked to rate an hour, people do not average the sixty minutes. They recall the most intense moment and the final one, and score roughly the midpoint of those two. The duration barely registers. In the original experiments a longer unpleasant episode was remembered as better than a shorter one, purely because it happened to finish more gently.
This is why the last thirty seconds of anything carry weight out of all proportion to their length, and why a single bright moment can hold up a whole mediocre stretch. Memory is not a recording. It is a headline and a closing line.
Asymmetry
Bad is stronger than good, and not by a little.
Losses are felt roughly twice as heavily as equivalent gains. A criticism outweighs a compliment; one sour note in an otherwise fine day is what gets retold at dinner. Across decades of research the pattern is stubborn enough that psychologists gave it a blunt name: bad is stronger than good.
There is a reason that is not pessimism. An ancestor who ignored a good berry lost a snack; one who ignored a bad sound lost everything. Attention that leans toward threat survives more often, so the mind we inherited is not a fair scale. It is an alarm that also happens to keep accounts.
Anchoring
The first number you hear bends every number after it.
Show someone a wheel of fortune, let it stop on a high figure, then ask an unrelated factual question with a numeric answer. Their estimate moves toward the number on the wheel. They know the wheel is random. It moves anyway.
Judgement does not start from nothing and work upward; it starts from whatever is nearest and adjusts, and it nearly always stops adjusting too early. The first price, the first opinion overheard, the first example encountered becomes the floor the mind negotiates against, long after the person has forgotten where it came from.
Other people
Knowing what others thought changes what you saw.
In Asch's line experiments, people denied the plain evidence of their eyes when enough others in the room did so first. Later work found something stranger still: sometimes the dissent is not social cowardice at all. Shown a group's answer before judging, the brain's own visual processing shifts. The person is not lying to fit in. They are seeing differently.
Consensus is not merely persuasive after the fact; it is an ingredient of perception itself. This is the uncomfortable heart of every aggregate opinion, and it is why the order in which judgements arrive shapes the judgements that follow.
Reconstruction
Every recollection is a rebuild, and the rebuild can be edited.
Memory does not retrieve a stored file; it reassembles one from fragments, filling gaps with what is plausible. Which means the act of remembering can itself introduce material. Ask witnesses how fast cars were going when they smashed rather than hit, and they report higher speeds, and are likelier to recall broken glass that was never there.
So a judgement formed weeks after the fact is not a faded copy of the original. It is a new object, built partly from the event and partly from every question, conversation and mood that came between. This is not a defect in people. It is the price of a memory that generalises instead of merely storing, and generalising is the thing that makes us able to think at all.
Finding
None of this makes human judgement worthless. It makes it human: fast, economical, built for a world that punishes hesitation. A rating was never a measurement in the first place. It is a memory of a feeling, reconstructed later, under the influence of whatever else was in the room. Read at that scale, and in numbers large enough for the noise to cancel, it is still the most honest signal we have about how it actually felt to be there.
Interlude
How a bazaar teaches you to buy
Shopping is learned behaviour, and different places teach it differently. What follows sets five habits common in Kurdistan beside five common in the United States. Not to rank them: each is a sensible answer to a different question, and each carries costs the other avoids.
Neither column is better. A fixed price buys speed and scale; a negotiated one buys judgement and relationship. A stranger's review buys reach; a cousin's recommendation buys accountability. The systems are trading different things, and both of them work.
A price is the start of a conversation, not the end of one
Negotiation as respect
In Kurdistan
The number on the box is an opening position, and testing it is not rudeness but participation. To pay the first figure without a word can read as disinterest, in the goods and in the seller. The exchange is part of the purchase: it establishes that both people are paying attention. What makes a customer angry is therefore rarely a high price. It is a price that moved after it was agreed, which converts a negotiation into a trick.
In the United States
The label is the price, and haggling over it in a shop would be mildly embarrassing for everyone. Fixed pricing was a nineteenth-century retail invention, and it bought something real: you can staff a thousand stores with clerks who need no authority to decide. The cost is that the price arrives as a decree, and the customer's only vote is to walk out.
You buy from a person, not from a shop
The named specialist
In Kurdistan
The barber, the goldsmith, the dentist are known by name, and that name travels through families. A reputation is held by an individual rather than a brand, and it is enforced socially: the man who cheats someone's cousin will hear about it at a wedding. This is why a shop can prosper for thirty years with no sign worth reading. The signage is other people's mouths.
In the United States
Trust is more often placed in the institution than the individual: the chain, the warranty, the return policy, the star rating of a stranger. It scales beautifully, letting you buy confidently in a city where you know nobody. The trade-off is that nobody in the transaction is personally accountable to you, and everyone involved is replaceable, including you.
Hospitality comes before business, and is not a technique
Tea first
In Kurdistan
Tea arrives before the discussion, and refusing it is harder than refusing the goods. A guest carries standing, so how a person is received is not a preliminary to the deal, it is a statement about who they are taken to be. This is the deep reason dismissiveness costs so much here: to be treated as an interruption is not poor service, it is a denial of standing, and people do not forgive that at the price they forgive a bad product.
In the United States
Warmth is real but professionalised, and its highest form is often efficiency: not wasting your time is the compliment. Friendliness is part of the job description rather than a personal obligation, which makes it reliable and portable, and also makes it something a customer can reasonably suspect of being a script.
The buyer is usually a family, even when one person is standing there
The collective decision
In Kurdistan
Significant purchases are discussed: parents, siblings, an uncle who knows about cars. The person at the counter may be carrying a decision they do not solely own, which is why they leave to consult and return days later. Advice moves along family lines faster and with more force than any advertisement, and a shop that satisfies one household has often acquired several.
In the United States
The default unit is the individual consumer, and choosing alone is treated as a mark of adulthood. Deciding fast is easy and the regret is private. What is lost is the friction that used to catch bad purchases early, which is partly why so much energy now goes into reviews: they are a manufactured substitute for the relative who would have told you not to.
The guarantee is the relationship, not the paperwork
Trust without a receipt
In Kurdistan
Much still runs on cash, memory and word given. If something breaks you return to the man who sold it, and the obligation is personal rather than contractual. It is flexible, fast and almost free to operate. It also means a stranger to the network carries real risk, and that the whole arrangement rests on being known, which not everyone equally is.
In the United States
Protection is externalised into systems: receipts, warranties, card chargebacks, thirty-day returns. A stranger can be trusted because the institution stands behind them, which is a genuine achievement of scale. The cost is a paperwork layer everyone pays for, and a habit of reading the terms instead of reading the person.
Finding
Read together, these five explain a pattern that runs through this whole report. Where trust is personal and prices are negotiated, the unforgivable failures are not defects but breaches: the price that changed, the promise that did not hold, the customer treated as an inconvenience. That is not a market behaving unusually. It is a market where the guarantee was never a document in the first place, so the only thing left to break is your word.
Chapter four
Every city grades differently
Select a city to see its profile. The differences between them are consistent enough that a single national ranking would be misleading.
For example
.
The smaller the town, the higher the score
Correlation
Each circle is a city, sized by how many businesses it holds. Horizontal is how much public attention an average business there receives; vertical is the average rating.
Average ratings track town size almost perfectly, and they run in the opposite direction to what you would expect. The smallest towns give the highest scores; the largest cities give the lowest. Quality is not what is being measured here, social distance is. Where a reviewer is likely to know the owner personally, an anonymous complaint is not really anonymous.
This is why a naive national leaderboard would be topped almost entirely by businesses in the smallest towns. Any honest ranking has to compare a business against its own city, not against the country.
Finding
Duhok is the most informative city in the dataset. It has the lowest average and the highest share of negative reviews among the three large cities, which means its customers are the most willing to say what went wrong.
Chapter five
A profile of every city
The same five sectors measured in each city, alongside what its reviewers talk about and what they are called. Cities are not smaller or larger versions of each other.
For example
.
Finding
.ten of the eleven cities where it can be measured, and restaurants are close to the worst in nearly all of them. People rate the thing they researched for months above the thing they chose in ten minutes.
Finding
How much a city talks varies more than what it says. Service comes up in 30.7% of reviews in Koya but only 5.8%.
Finding
Reviewer names split cleanly by city size. The three metros are led by Ahmed, Ali and Mohammed. The towns surface distinctly Kurdish names instead: Nechirvan in Zakho, Rekawt in Chamchamal, Ramyar, Hiwa and Aram in Ranya, Diyar in Kalar.
What we are building
The same corpus, turned into decisions
A report tells you what happened. These tools are the working end of the same data: they answer the question a specific person is actually asking, on a specific street, about a specific business.
Live on the business dashboard
Canney AI
Fed the whole market, so it understands your street better than any one owner can.
Canney AI reads across the whole corpus at once: what customers praise and complain about in your category, how the businesses on your street score, what service standard your neighbourhood has quietly settled on. For an owner it turns a few hundred thousand scattered opinions into a short answer to a plain question, whether that is what to fix first, what a fair price looks like here, or where demand is going unserved.
Reads your reviews, your category and your city as three separate signals
Benchmarks you against businesses like yours, not against the whole market
Learns continuously: every new review and post sharpens what it knows about your market
An agent you can talk to, answering questions around the clock
In development
Site Intelligence
Give it an address. It tells you what that ground is worth trading on.
Before signing a lease or building on a plot, the questions are always the same and are usually answered by rumour: who is already here, how far is the nearest competitor, is this street underserved or saturated, what do customers in this neighbourhood reward and what do they complain about. Site Intelligence answers them from the corpus instead. You enter an address and it reads the businesses around that point, how they score, how far apart they sit, and what the area is measurably short of.
Every business within walking distance, by category and rating
Distance to the nearest direct competitor, and how many sit inside the same radius
The neighbourhood benchmark: what a good rating means on this street, not citywide
Gaps: the categories this area is short of, measured against comparable streets
Chapter six
A respect economy
Set aside the ratings and look only at the words. The vocabulary customers reach for when they are angry says something specific about this market.
For example
.
The words that separate praise from anger
Terms appearing far more often in one sentiment than the other, across the reviews written in English.
Appearances per 10,000 English reviews of that sentiment. The multiplier: how many times more likely the word is there than in the opposite sentiment.
The complaint vocabulary is dominated by words about how a person was treated: unprofessional, disrespectful, arrogant, rude, ignored. Words about broken or defective products barely register by comparison. The praise vocabulary mirrors it exactly: knowledgeable, responsive, elegant, a pleasure.
There is a second pattern worth naming. Negative reviews in this corpus run roughly twice as long as positive ones. Praise is a gesture; criticism is an argument. Nearly a tenth of all reviews also invoke the city or the country by name, and those reviews rate well above average. Leaving a review here is partly an act of local endorsement, not only a consumer verdict.
Finding
A business in Kurdistan loses stars for making a customer feel small far more reliably than for selling a mediocre product. Staff manner is not a soft metric here. It is the primary one.
Chapter seven
Where attention is moving
Comparing how much review activity each category attracts now against a few years ago, indexed so the market as a whole sits at 100, shows a clear directional shift.
For example
.
Attention index by category
100 means a category is growing exactly as fast as the market. Above 100 it is gaining share of attention; below 100 it is losing it. Categories with too little history to compare are not shown.
Gaining shareLosing share
Attention against satisfaction
Sector matrix
Every category, placed by how many reviews the average business attracts and how well it scores. Circle size is how many businesses are in the category. The busiest categories sit lowest.
Everything gaining is a booked, personal service: barbers, pharmacies, dental clinics, salons, coffee. Everything losing is general provisioning: the bazaar, the shopping mall, the supermarket, the grocery, the fuel station.
That is the signature of a market where discretionary spending is rising and moving off the general marketplace onto specialised services. It is also a shift from places you pass through weekly to places you make an appointment with, which changes what a directory has to do for them.
Finding
The mall era is visibly maturing. Shopping malls attract more reviews per business than any other category in the country and are simultaneously among the weakest on new attention: a format with enormous accumulated audience and slowing momentum.
Chapter eight
Six things the averages hide
Headline numbers flatten a market. These five cuts of the same data each contradict something the averages imply.
For example
.
The more reviews a business has, the worse it scores
Popularity
Businesses grouped by how many reviews they carry. Height is the average rating of that group; the number under each column is how many businesses sit in it.
Finding
A business with under five reviews averages 4.57. One with fifty to a hundred averages 4.07. Half a star vanishes purely by becoming known. Early reviews come from people who already like you; scale brings strangers. Never compare a new listing's rating to an established one.
Experienced reviewers are tougher, but less extreme
Reviewers
Reviewers grouped by how many reviews they have written. Both their average rating and their rate of one-star reviews fall as they write more.
Average rating givenShare of one-star reviews
Finding
First-time reviewers give 4.38 and leave one star 9.3% of the time. People with 25 or more reviews give 3.97 and leave one star only 4.9% of the time. Experience makes a reviewer stingier with fives and slower to rage. The harshest voices in this corpus are not the angriest ones.
Which trades cluster, and which spread out
Geography
Median walking distance from a business to the nearest other business of the same kind. Short bars are trades that pile onto one street; long bars are services that spread across neighbourhoods.
Finding
.
The spread inside a category matters more than its average
Variance
Each row is a category. The bar spans the middle half of its businesses; the line runs from worst to best; the marker is the median business. A wide bar means choosing well matters.
Finding
Two categories can share an average and behave nothing alike. Where the middle half is tight, almost any choice is fine. Where it is wide, the gap between the typical business and a bad one is large enough that a recommendation is worth real money. Variance, not average, is what a directory is actually for.
Where every city concentrates its businesses
Matrix
Share of each city's listings by category. Read down a column to compare cities on one trade; read across a row to see what a city is made of.
How much people bother to write
Effort
Median characters per review, by category. Effort tracks how consequential the decision felt, not how much the customer spent.
Finding
People write essays about clinics and one word about petrol. Effort follows risk, not price: the decisions people explain at length are the ones that were hard to reverse. Categories at the top of this chart are where written reviews carry real information; at the bottom, a star rating is all you will ever get.
Chapter nine
The conversation almost nobody is having
Customers here write a great deal. Businesses answer very little, and most listings are missing the basic facts a customer needs before walking in.
For example
.
How complete a typical listing is
Share of all businesses that publish each detail.
Which sectors reply to their customers
Share of reviews that received a reply from the business.
The overwhelming majority of reviews in Kurdistan are never answered by anyone. The sectors that do reply are the internationally run ones: the Lebanese and Italian restaurants, the travel agencies, the electronics chains. Replying is a learned habit, not a cultural absence, and almost nobody here has learned it yet.
The gap in listing detail is larger still. Most businesses in Kurdistan publish no website at all, and roughly two thirds publish no opening hours, which is the single most useful fact a directory can carry. This is not a flaw in the market. It is the opportunity: for most businesses here, a good directory listing is not a supplement to their web presence, it is the whole of it.
Finding
The reply layer is effectively empty, and the businesses already using it prove there is demand. Answering reviews is the cheapest available advantage in this market, and it is currently unclaimed.
Industry insight
Every industry, read on its own terms
The charts above compare categories against each other. This reads each one on its own: how big it is, how it scores, which way it is moving, and the single change most likely to move its rating.
Second edition
We collected more, and ran the whole analysis again
This report is not written once. Every figure on this page is recomputed from the full corpus, so when the corpus grows the report is rebuilt rather than patched. Here is what the second pass changed, and what it left standing.
In short
Six things worth remembering
01
Price is what decides a rating. Service is what people talk about.
Food and service lead the conversation, yet both rate near the market line. Price is raised far less often, and it is the one that decides the score. In a bazaar culture where a price is a conversation, a fixed number that feels unfair reads as a breach of trust, not a market rate. Compete on how the visit felt; defend the rating by never letting a price surprise anyone at the till.
02
A broken promise costs far more stars than an ordinary disappointment.
The angriest themes in the corpus are all promises: the Wi-Fi that does not connect, the delivery that arrives late, the quoted price that grows. A plain kebab shop with honest portions outscores a lavish venue that overpromises. Guests here forgive modest; they do not forgive misled.
03
Customers punish disrespect harder than they punish defects.
The one-star vocabulary is about being made small: rude, ignored, arrogant, unprofessional. In a society where a guest expects honour at the door, a dismissive clerk costs more than a faulty product, and reviews that mention hospitality or feeling welcomed sit far above the market average. Train the greeting before the workflow.
04
Attention is moving from general marketplaces to booked personal services.
Barbers, beauty salons, dental clinics and pharmacies are gaining review share fastest while general retail fades. Kurdistan increasingly follows a person, not a shelf: customers book a specific barber or doctor by name and the review is part of choosing them. A named specialist with a full profile is the strongest position on the platform.
05
Small towns rate higher than large cities, so rank within a city, never across the country.
Ranya and Chamchamal average well above Erbil, and it is not because the shops are better. In a town where the reviewer may greet the owner's family in the market, a public one-star is a social act, not an anonymous verdict. A 4.3 in Erbil can be a stronger signal than a 4.6 in a small town; always compare a business with its own street.
06
Almost nobody replies to reviews, and most listings are missing their opening hours.
About 94 in 100 reviews never receive an answer, and {h} listings in ten publish no opening hours. In a market that runs on word of mouth and family recommendation, the first business that answers politely, in Kurdish, with a fix, looks remarkable to more than a hundred thousand review writers. The bar for standing out has never been lower.
Own your listing
Add your opening hours, answer your reviews, and show customers the two things this data says they care about most. It is free.
This report reads the complete Kurdify corpus, not a sample. Ratings are averaged only over reviews that carry a star rating; reviews without one are excluded from every average. Theme percentages are calculated over reviews that contain text, since a rating-only review has nothing to categorise.
Themes are matched with a keyword lexicon rather than a language model, so a review counts towards a theme when it uses that theme's vocabulary. The attention index compares recent review volume against earlier volume and is indexed to the market as a whole, which cancels out the general growth of the platform. Categories and cities below a minimum size are not charted.
